
# установка пакета для чтения файлов в формате xls
# pip install openpyxl


# импорт библиотек для работы
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt

from statsmodels.tsa.seasonal import seasonal_decompose
import time

import lightgbm as lgb
from sklearn.model_selection import train_test_split
from sklearn.model_selection import TimeSeriesSplit
from catboost import CatBoostRegressor
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error, r2_score, mean_absolute_error, mean_absolute_percentage_error
from sklearn.model_selection import GridSearchCV
from catboost import Pool, cv
from sklearn.metrics import accuracy_score
from scipy.signal import find_peaks

import optuna
from optuna.visualization import plot_optimization_history, plot_param_importances

pd.set_option('display.max_columns', None)

import warnings
warnings.filterwarnings('ignore')


# загрузка данных из файла
df = pd.read_excel('array.xlsx', index_col=[0], parse_dates=[0]) 
# сортировка индексов
df.sort_index(inplace=True)

print(f'Загружены данные за период {min(df.index)} - {max(df.index)}')

Загружены данные за период 2020-01-03 00:00:00 - 2022-08-28 00:00:00


# выведем последние 5 строк
df.tail()


# вывод общей информации для проверки типов данных
df.info()

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 965 entries, 2020-01-03 to 2022-08-28
Data columns (total 5 columns):
 #   Column   Non-Null Count  Dtype  
---  ------   --------------  -----  
 0   picking  965 non-null    int64  
 1   amount   965 non-null    int64  
 2   income   965 non-null    float64
 3   loads    965 non-null    int64  
 4   holiday  965 non-null    int64  
dtypes: float64(1), int64(4)
memory usage: 45.2 KB


# миллионы сократим на 1000
df['income'] = df['income']/1000


# общая информация о данных
df.describe()


for col in df.columns[:-1]:
    plt.figure(figsize=(12,4))
    ax = sns.boxplot(x=col, data=df, color="green")
    
    ax = sns.stripplot(x=col, data=df, color="orange", jitter=1, size=2.5)
    plt.title(f'«Ящик с усами» по {col}')
    plt.xlabel(f'Кол-во {col}');


# напишем функцию для построения графиков
def plot_rolling_mean(df, win_n):
    for col in df.columns[:-1]:
        rolling_mean = df[col].rolling(window=win_n).mean()
        plt.figure(figsize=(16,6))
        plt.title(f'{col} со скользящим средним = {win_n} дней')
        plt.plot(rolling_mean, "g", label="Скользящее среднее")
        plt.plot(df[col][win_n:], label=f"Кол-во {col}", alpha=.5)
        plt.legend(loc="upper left")


plot_rolling_mean(df, 7)


plot_rolling_mean(df, 30)


# раскладываем временной ряд на три составляющие: тренд, сезонность и остаток (шум)
def decomposed(df, col):
    decomposed = seasonal_decompose(df['2020':'2021'][col], extrapolate_trend='freq', period=30) 
    
    # строим графики
    plt.figure(figsize=(12, 10))
    plt.subplot(311)
    decomposed.trend.plot(ax=plt.gca())
    plt.title(f'Тренд {col}')
    plt.xlabel('период')
    
    plt.subplot(312)
    decomposed.seasonal.plot(ax=plt.gca())
    plt.title(f'Сезонность {col}')
    plt.xlabel('период')
    
    plt.subplot(313)
    decomposed.resid.plot(ax=plt.gca()) 
    plt.title(f'Шум {col}')
    plt.xlabel('период')
    plt.tight_layout()


decomposed(df, 'picking')


decomposed(df, 'income')


decomposed = seasonal_decompose(df['2022-06-01':'2022-08-28']['picking'], extrapolate_trend='freq', period=30) 
decomposed.seasonal.tail(8).plot(title='Сезонность понедельно', figsize=(12, 5))
plt.title('Сезонность внутри недели')
plt.xlabel('День месяца')
plt.show();


df['mean'] = df['picking'].rolling(7).mean()
df['std'] = df['picking'].rolling(7).std()
df[['mean', 'std']]['2022'].plot(kind='line', style='-', figsize=(16,6), rot=45)
plt.title('Среднеее и стандартное отклонение')
plt.xlabel('Период');


# сделаем копию датасета, чтобы не перезаписывать исхдные данные
df_copy = df[['picking']].copy()
plt.figure(figsize=(16, 8))
# вычисляем разность ряда
df_copy['picking'] = df_copy - df_copy.shift()
df_copy['mean'] = df_copy['picking'].rolling(7).mean()
df_copy['std'] = df_copy['picking'].rolling(7).std()
df_copy['2021':'2022'].plot(ax=plt.gca(), alpha=.42)
plt.title('Разница ряда')
plt.xlabel('Период')
plt.show();


# добавляем признаки по дате
df['month'] = df.index.month
df['day'] = df.index.day
df['dayofweek'] = df.index.dayofweek

df.tail()


# добавляем признаки со смещенеием занчений
for lag in range(1, 10):
    df['lag_{}'.format(lag)] = df['mean'].shift(lag)
# признак скользящего среднего по среднему    
df['rolling_mean'] = df['mean'].shift().rolling(7).mean()
# признак сезонности
df['seasonal'] = seasonal_decompose(df['picking'], extrapolate_trend='freq', period=30).seasonal
# признак тренда
df['trend'] = seasonal_decompose(df['picking'], extrapolate_trend='freq', period=30).trend
# признак скользящего окна (среднее)
df['expanding_window'] = df['picking'].expanding(7).mean()
# признак по кол-ву пиков
df['kurtosis'] = df['picking'].rolling(6).apply(lambda x: len(find_peaks(x)[0])) 
# признаки по разности рядов
for d in range(1,8):
    df['diff_{}'.format(d)] = df['picking'].diff(periods=d)
df.tail(8)


# удалим нулевые значения
df = df.dropna()


random_state = 12345


# сделаем копию нашего датасета
data = df.copy()
target = data['picking']
features = data.drop(['picking', 'income', 'amount', 'loads'], axis=1)
X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, shuffle=False)


def objective(trial):
    
    params = {
        'metric': 'rmse', 
        'random_state': random_state,
        'n_estimators': 10000,
        'reg_alpha': trial.suggest_loguniform('reg_alpha', 1e-3, 10.0),
        'reg_lambda': trial.suggest_loguniform('reg_lambda', 1e-3, 10.0),
        'colsample_bytree': trial.suggest_categorical('colsample_bytree', [0.3,0.35,0.4,0.45,0.5,0.6,0.7,0.75,0.8,0.85]),
        'subsample': trial.suggest_categorical('subsample', [0.6,0.65,0.7,0.75,0.8,0.85]),
        'learning_rate': trial.suggest_categorical('learning_rate', 
                                                   [0.005,0.006,0.008,0.01,0.015,0.02,0.03]),
        'max_depth': trial.suggest_categorical('max_depth', [-1,10,20]),
        'num_leaves' : trial.suggest_int('num_leaves', 1, 300),
        'min_child_samples': trial.suggest_int('min_child_samples', 1, 300),
        'cat_smooth' : trial.suggest_int('min_data_per_groups', 1, 100)
    }
    
    # Создаем и обучаем модель с мониторингом переобучения
    model = lgb.LGBMRegressor(**params)      
    model.fit(X_train, y_train, eval_set=[(X_test,y_test)], early_stopping_rounds=300, verbose=False)
    preds = model.predict(X_test)
    
    #Смотрим финальную метрику RMSE
    rmse = mean_squared_error(y_test, preds, squared=False)
    
    return rmse


%%time
# Посчитаем время оптимизации

# Создаем задание для фреймворка
study = optuna.create_study(direction='minimize')  # Минимизируем ошибку
study.optimize(objective, n_trials=20)              # Количество итераций = 20

# Смотрим на финальные метрики
print('Number of finished trials:', len(study.trials))
print('Best trial:', study.best_trial.params)
print('Best score:', study.best_trial.value)

[I 2022-08-29 16:05:01,429] A new study created in memory with name: no-name-caecb9ed-7a23-483d-b03d-19ae6cc6cea5
[I 2022-08-29 16:05:02,277] Trial 0 finished with value: 2372.2796437730094 and parameters: {'reg_alpha': 4.706339674385066, 'reg_lambda': 0.0016137227141746896, 'colsample_bytree': 0.6, 'subsample': 0.7, 'learning_rate': 0.03, 'max_depth': 20, 'num_leaves': 90, 'min_child_samples': 38, 'min_data_per_groups': 25}. Best is trial 0 with value: 2372.2796437730094.
[I 2022-08-29 16:05:09,253] Trial 1 finished with value: 1849.6762016517284 and parameters: {'reg_alpha': 1.6512320187036338, 'reg_lambda': 0.01531381732317706, 'colsample_bytree': 0.3, 'subsample': 0.6, 'learning_rate': 0.03, 'max_depth': -1, 'num_leaves': 149, 'min_child_samples': 4, 'min_data_per_groups': 45}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:10,760] Trial 2 finished with value: 2713.2404140063704 and parameters: {'reg_alpha': 0.008105296686393995, 'reg_lambda': 0.0012779742599202339, 'colsample_bytree': 0.6, 'subsample': 0.8, 'learning_rate': 0.008, 'max_depth': 10, 'num_leaves': 142, 'min_child_samples': 158, 'min_data_per_groups': 20}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:18,787] Trial 3 finished with value: 1874.8650266059271 and parameters: {'reg_alpha': 0.6461618668078962, 'reg_lambda': 7.333415258575352, 'colsample_bytree': 0.5, 'subsample': 0.65, 'learning_rate': 0.03, 'max_depth': 10, 'num_leaves': 82, 'min_child_samples': 5, 'min_data_per_groups': 51}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:19,994] Trial 4 finished with value: 2641.443748914188 and parameters: {'reg_alpha': 0.003459470853855497, 'reg_lambda': 0.7167647924725716, 'colsample_bytree': 0.5, 'subsample': 0.7, 'learning_rate': 0.01, 'max_depth': 20, 'num_leaves': 201, 'min_child_samples': 132, 'min_data_per_groups': 29}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:22,952] Trial 5 finished with value: 2510.4546689164417 and parameters: {'reg_alpha': 0.0035760960063698746, 'reg_lambda': 0.5189242855193567, 'colsample_bytree': 0.45, 'subsample': 0.65, 'learning_rate': 0.006, 'max_depth': 10, 'num_leaves': 155, 'min_child_samples': 103, 'min_data_per_groups': 71}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:23,203] Trial 6 finished with value: 5437.945629740844 and parameters: {'reg_alpha': 0.7347368156927587, 'reg_lambda': 0.0712318987673358, 'colsample_bytree': 0.4, 'subsample': 0.7, 'learning_rate': 0.006, 'max_depth': -1, 'num_leaves': 77, 'min_child_samples': 274, 'min_data_per_groups': 96}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:23,735] Trial 7 finished with value: 2585.7581797047897 and parameters: {'reg_alpha': 0.0219421316234564, 'reg_lambda': 0.009103585100192595, 'colsample_bytree': 0.85, 'subsample': 0.75, 'learning_rate': 0.02, 'max_depth': 10, 'num_leaves': 25, 'min_child_samples': 129, 'min_data_per_groups': 79}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:23,837] Trial 8 finished with value: 5685.002635966603 and parameters: {'reg_alpha': 0.5388250448804189, 'reg_lambda': 0.013296734935835755, 'colsample_bytree': 0.6, 'subsample': 0.6, 'learning_rate': 0.015, 'max_depth': 20, 'num_leaves': 64, 'min_child_samples': 297, 'min_data_per_groups': 88}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:24,365] Trial 9 finished with value: 2225.4496141028017 and parameters: {'reg_alpha': 6.913824921232273, 'reg_lambda': 0.01799370968395574, 'colsample_bytree': 0.45, 'subsample': 0.7, 'learning_rate': 0.02, 'max_depth': 20, 'num_leaves': 300, 'min_child_samples': 43, 'min_data_per_groups': 98}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:24,710] Trial 10 finished with value: 4171.823745844223 and parameters: {'reg_alpha': 0.08104378085345017, 'reg_lambda': 0.15016215830682614, 'colsample_bytree': 0.3, 'subsample': 0.6, 'learning_rate': 0.005, 'max_depth': -1, 'num_leaves': 244, 'min_child_samples': 210, 'min_data_per_groups': 51}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:33,806] Trial 11 finished with value: 2099.0501407244615 and parameters: {'reg_alpha': 0.7221022572603432, 'reg_lambda': 6.7894877653027015, 'colsample_bytree': 0.35, 'subsample': 0.85, 'learning_rate': 0.03, 'max_depth': -1, 'num_leaves': 138, 'min_child_samples': 2, 'min_data_per_groups': 49}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:51,136] Trial 12 finished with value: 2683.908958687852 and parameters: {'reg_alpha': 0.2389583409819274, 'reg_lambda': 7.8194517265511525, 'colsample_bytree': 0.8, 'subsample': 0.65, 'learning_rate': 0.03, 'max_depth': -1, 'num_leaves': 110, 'min_child_samples': 1, 'min_data_per_groups': 1}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:51,906] Trial 13 finished with value: 2300.477630770995 and parameters: {'reg_alpha': 2.141783937686881, 'reg_lambda': 1.6775020679673214, 'colsample_bytree': 0.3, 'subsample': 0.65, 'learning_rate': 0.03, 'max_depth': 10, 'num_leaves': 195, 'min_child_samples': 70, 'min_data_per_groups': 55}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:52,434] Trial 14 finished with value: 2449.821901140701 and parameters: {'reg_alpha': 0.08101750618351303, 'reg_lambda': 0.06747519011367267, 'colsample_bytree': 0.5, 'subsample': 0.6, 'learning_rate': 0.03, 'max_depth': -1, 'num_leaves': 21, 'min_child_samples': 71, 'min_data_per_groups': 41}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:05:56,939] Trial 15 finished with value: 3081.6140055400124 and parameters: {'reg_alpha': 1.9252776192469716, 'reg_lambda': 0.19462073914417166, 'colsample_bytree': 0.75, 'subsample': 0.75, 'learning_rate': 0.005, 'max_depth': 10, 'num_leaves': 179, 'min_child_samples': 190, 'min_data_per_groups': 67}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:06:01,753] Trial 16 finished with value: 2218.5895076297775 and parameters: {'reg_alpha': 0.21666306161715462, 'reg_lambda': 0.005922074997983978, 'colsample_bytree': 0.7, 'subsample': 0.85, 'learning_rate': 0.008, 'max_depth': -1, 'num_leaves': 48, 'min_child_samples': 34, 'min_data_per_groups': 37}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:06:02,355] Trial 17 finished with value: 2387.720132752521 and parameters: {'reg_alpha': 2.341714511074219, 'reg_lambda': 0.02947034570371361, 'colsample_bytree': 0.3, 'subsample': 0.8, 'learning_rate': 0.015, 'max_depth': 10, 'num_leaves': 117, 'min_child_samples': 87, 'min_data_per_groups': 63}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:06:04,335] Trial 18 finished with value: 1890.061635091286 and parameters: {'reg_alpha': 0.0010215440169948802, 'reg_lambda': 1.4504681673039455, 'colsample_bytree': 0.5, 'subsample': 0.65, 'learning_rate': 0.01, 'max_depth': -1, 'num_leaves': 213, 'min_child_samples': 25, 'min_data_per_groups': 8}. Best is trial 1 with value: 1849.6762016517284.
[I 2022-08-29 16:06:06,359] Trial 19 finished with value: 3528.4070194056912 and parameters: {'reg_alpha': 0.2741378444278379, 'reg_lambda': 0.003648295570053628, 'colsample_bytree': 0.7, 'subsample': 0.6, 'learning_rate': 0.03, 'max_depth': 10, 'num_leaves': 249, 'min_child_samples': 225, 'min_data_per_groups': 38}. Best is trial 1 with value: 1849.6762016517284.

Number of finished trials: 20
Best trial: {'reg_alpha': 1.6512320187036338, 'reg_lambda': 0.01531381732317706, 'colsample_bytree': 0.3, 'subsample': 0.6, 'learning_rate': 0.03, 'max_depth': -1, 'num_leaves': 149, 'min_child_samples': 4, 'min_data_per_groups': 45}
Best score: 1849.6762016517284
CPU times: total: 1min 28s
Wall time: 1min 4s


# История оптимизации
plot_optimization_history(study)


# Значимость гиперпараметров при настройке
plot_param_importances(study)


# Выводим лучшие гиперпараметры
study.best_params

{'reg_alpha': 1.6512320187036338,
 'reg_lambda': 0.01531381732317706,
 'colsample_bytree': 0.3,
 'subsample': 0.6,
 'learning_rate': 0.03,
 'max_depth': -1,
 'num_leaves': 149,
 'min_child_samples': 4,
 'min_data_per_groups': 45}


optuna_params = study.best_params

optuna_params['metric'] = 'rmse'
# optuna_params['random_state'] = random_state
optuna_params['n_estimators'] = 10000


# Создаем и обучаем модель с подобранными гиперпараметрами
model_optuna = lgb.LGBMRegressor(**optuna_params)
model_optuna.fit(X_train, y_train, eval_set=[(X_test, y_test)], early_stopping_rounds=300, verbose=300)

[LightGBM] [Warning] Unknown parameter: min_data_per_groups
[300]	valid_0's rmse: 1869.12
[600]	valid_0's rmse: 1867.44

LGBMRegressor(colsample_bytree=0.3, learning_rate=0.03, metric='rmse',
              min_child_samples=4, min_data_per_groups=45, n_estimators=10000,
              num_leaves=149, reg_alpha=1.6512320187036338,
              reg_lambda=0.01531381732317706, subsample=0.6)


%%time
start_time = time.time()

pred_optuna = model_optuna.predict(X_test)
rmse_optuna = mean_squared_error(y_test, pred_optuna, squared=False)

end_time = time.time()
time_tst = end_time - start_time

print('RMSE модели c подбором параметров через OPTUNA:', rmse_optuna)
print('Время теста c подбором параметров через OPTUNA:', time_tst)

RMSE модели c подбором параметров через OPTUNA: 1866.5839634472334
Время теста c подбором параметров через OPTUNA: 0.02727961540222168
CPU times: total: 15.6 ms
Wall time: 27.3 ms


# важность признаков LightGBM
lgb.plot_importance(model_optuna, figsize = (18, 10))
plt.show()


a = pd.DataFrame(y_test)
a['pred_optuna'] = pred_optuna
a = a.sort_index()


plt.figure(figsize=(16,6))
plt.plot(a.index, a['picking'], label='истинные значения')
plt.plot(a.index, a['pred_optuna'], label='предсказания')
plt.title('Истинные значения vs предсказания')
plt.legend()
plt.ylim(0)
plt.show()


print('MAE:', mean_absolute_error(y_test, pred_optuna))
print('R2:', r2_score(y_test, pred_optuna))
print('MAPE:', mean_absolute_percentage_error(y_test, pred_optuna))

MAE: 1339.513572124153
R2: 0.969662660999613
MAPE: 0.08994134576248568


a.tail(15)

	picking	amount	income	loads	holiday
date
2022-08-24	32667	119868	30368352.50	20344	0
2022-08-25	31435	99598	28055369.18	19480	0
2022-08-26	30333	103285	27522253.52	19513	0
2022-08-27	7299	18296	7208401.35	5023	0
2022-08-28	16363	47394	14541160.30	9123	0

	picking	amount	income	loads	holiday
count	965.000000	965.000000	965.000000	965.000000	965.000000
mean	28171.921244	93179.926425	22821.363300	15482.465285	0.079793
std	14530.773142	50529.902495	11559.120177	7512.889238	0.271113
min	1.000000	2.000000	0.466400	1.000000	0.000000
25%	20550.000000	63927.000000	16101.927540	11413.000000	0.000000
50%	30051.000000	101379.000000	25229.146680	17196.000000	0.000000
75%	36723.000000	122539.000000	29768.442760	20086.000000	0.000000
max	82379.000000	524976.000000	54132.616600	39041.000000	1.000000

	picking	amount	income	loads	holiday	mean	std	month	day	dayofweek
date
2022-08-24	32667	119868	30368.35250	20344	0	28147.857143	11666.774082	8	24	2
2022-08-25	31435	99598	28055.36918	19480	0	27763.142857	11479.611411	8	25	3
2022-08-26	30333	103285	27522.25352	19513	0	27589.285714	11421.771341	8	26	4
2022-08-27	7299	18296	7208.40135	5023	0	27669.000000	11252.713451	8	27	5
2022-08-28	16363	47394	14541.16030	9123	0	27435.857143	11500.726288	8	28	6

	picking	amount	income	loads	holiday	mean	std	month	day	dayofweek	lag_1	lag_2	lag_3	lag_4	lag_5	lag_6	lag_7	lag_8	lag_9	rolling_mean	seasonal	trend	expanding_window	kurtosis	diff_1	diff_2	diff_3	diff_4	diff_5	diff_6	diff_7
date
2022-08-21	17995	54808	16971.72571	10510	0	28596.142857	12333.411591	8	21	6	28475.142857	28654.142857	29163.714286	28775.000000	28761.000000	28309.428571	27669.000000	27464.000000	27326.714286	28543.918367	-183.525912	28296.298535	28177.299582	1.0	11254.0	-13555.0	-16133.0	-15222.0	-14162.0	-26390.0	847.0
2022-08-22	40957	132021	41012.95614	23848	0	28106.428571	11651.196560	8	22	0	28596.142857	28475.142857	28654.142857	29163.714286	28775.000000	28761.000000	28309.428571	27669.000000	27464.000000	28676.367347	-789.556063	28358.630033	28190.625652	1.0	22962.0	34216.0	9407.0	6829.0	7740.0	8800.0	-3428.0
2022-08-23	32997	133203	30388.71164	20701	0	28226.428571	11704.073705	8	23	1	28106.428571	28596.142857	28475.142857	28654.142857	29163.714286	28775.000000	28761.000000	28309.428571	27669.000000	28647.367347	247.804411	28420.961531	28195.632292	1.0	-7960.0	15002.0	26256.0	1447.0	-1131.0	-220.0	840.0
2022-08-24	32667	119868	30368.35250	20344	0	28147.857143	11666.774082	8	24	2	28226.428571	28106.428571	28596.142857	28475.142857	28654.142857	29163.714286	28775.000000	28761.000000	28309.428571	28571.000000	-815.494276	28483.293029	28200.285120	1.0	-330.0	-8290.0	14672.0	25926.0	1117.0	-1461.0	-550.0
2022-08-25	31435	99598	28055.36918	19480	0	27763.142857	11479.611411	8	25	3	28147.857143	28226.428571	28106.428571	28596.142857	28475.142857	28654.142857	29163.714286	28775.000000	28761.000000	28481.408163	-432.984943	28545.624527	28203.647609	1.0	-1232.0	-1562.0	-9522.0	13440.0	24694.0	-115.0	-2693.0
2022-08-26	30333	103285	27522.25352	19513	0	27589.285714	11421.771341	8	26	4	27763.142857	28147.857143	28226.428571	28106.428571	28596.142857	28475.142857	28654.142857	29163.714286	28775.000000	28281.326531	-1394.426620	28607.956025	28205.858775	1.0	-1102.0	-2334.0	-2664.0	-10624.0	12338.0	23592.0	-1217.0
2022-08-27	7299	18296	7208.40135	5023	0	27669.000000	11252.713451	8	27	5	27589.285714	27763.142857	28147.857143	28226.428571	28106.428571	28596.142857	28475.142857	28654.142857	29163.714286	28129.204082	-1295.817287	28670.287523	28184.171162	0.0	-23034.0	-24136.0	-25368.0	-25698.0	-33658.0	-10696.0	558.0
2022-08-28	16363	47394	14541.16030	9123	0	27435.857143	11500.726288	8	28	6	27669.000000	27589.285714	27763.142857	28147.857143	28226.428571	28106.428571	28596.142857	28475.142857	28654.142857	28014.040816	-675.719571	28732.619021	28171.921244	0.0	9064.0	-13970.0	-15072.0	-16304.0	-16634.0	-24594.0	-1632.0

	picking	pred_optuna
date
2022-08-14	17148	16736.476094
2022-08-15	44385	43113.266678
2022-08-16	32157	30016.305558
2022-08-17	33217	33228.619254
2022-08-18	34128	33397.645089
2022-08-19	31550	30953.234482
2022-08-20	6741	4811.592927
2022-08-21	17995	19275.996344
2022-08-22	40957	38687.299652
2022-08-23	32997	31813.687977
2022-08-24	32667	31738.759491
2022-08-25	31435	31616.976920
2022-08-26	30333	31786.858345
2022-08-27	7299	5417.908583
2022-08-28	16363	15148.776628

Прогноз объема работ по отгрузке на складах¶

Загрузка данных¶

Описание данных¶

Анализ данных¶

Скользящее среднее¶

Тренды и сезонность¶

Стационарные ряды и разности временного ряда¶

Обучение¶

Подготовка признаков¶

Обучение модели и подбор гипперпараметров¶